{
  "id": 14291848,
  "title": "How artificial neural networks build their own learning machinery",
  "url": "https://urgent.news/2026/10/13/how-artificial-neural-networks-build-their-own-learning-machinery",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-13T21:00:00.000Z",
  "source": {
    "name": "MIT News Research",
    "slug": "mit-news-research",
    "url": "https://news.mit.edu/2026/how-artificial-neural-networks-build-their-own-learning-machinery-1013"
  },
  "original_language": "en",
  "account": "In a recent study published in PNAS, scientists from MIT’s McGovern Institute for Brain Research and K. Lisa Yang Integrative Computational Neuroscience (ICoN) Center have demonstrated that artificial neural networks can develop their own learning mechanisms. These networks, which are capable of recognizing images, using language, or writing computer code, reorganize themselves as they learn, adjusting the connections between their components to enhance their performance. The researchers argue that a key factor for the biological relevance of an artificial neural network's learning process is its ability to self-organize, without the influence of a programmer. By implementing simple rules that govern how connections should change, a set of randomly connected units can transform into an effective information-processing network, learning from the data it is trained on. Current AI technology excels in certain tasks, but it lacks the learning algorithms found in the brain. Qianli Liao, a postdoctoral fellow at MIT, explains that while engineers design the rules governing connection changes during training, this approach may not provide insights into biological learning. The brain, unlike human-designed networks, organizes itself through unknown rules that could change over time or in different brain regions. Understanding these rules could lead to breakthroughs in therapies for age-related or disease-induced impairments, as well as the development of advanced brain-computer interfaces. The researchers propose that the value of an artificial neural network to neuroscience should be assessed based on its capacity to self-organize, rather than just its connectivity. In their study, they developed a self-organizing computational network that learns to recognize patterns and classify images using simple rules similar to those thought to underlie brain circuit rewiring. These rules include \"neurons that fire together wire together,\" meaning that connections between active neurons strengthen over time. The team's model employs an optimization strategy akin to gradient descent, a widely used machine learning algorithm used by most modern AI systems. However, this approach differs from the back-propagation algorithm, which neuroscientists believe is unlikely to be employed by the brain. By comparing their self-organizing network to other networks designed to learn in a similar manner, the researchers found that it performed comparably to others. This breakthrough could pave the way for significant progress in both AI and neuroscience, as finding a biological parallel to back-propagation could bridge the gap between artificial and natural networks.",
  "summary": "Machine-learning networks can learn without detailed instructions from a designer. Those that do may offer insight into how the brain reorganizes itself during learning.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "MIT News AI",
        "title": "How artificial neural networks build their own learning machinery",
        "url": "https://urgent.news/2026/10/13/how-artificial-neural-networks-build-their-own-learning-machinery-14296429",
        "published": "2026-10-13T21:00:00.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}